Tactile sensing provides valuable signals for robots to physically interact with their environment, resulting in more robust behavior when control and planning algorithms actively seek tactile measurements. However, it can be challenging to determine when and where to seek contact while balancing task fulfillment and uncertainty reduction.
Current gradient-based solutions may not work well for discrete domains like tactile sensing. We aim to develop a gradient-based control approach that considers the robot's belief to find good trade-offs for an example task, tactile localization. In tactile localization, the robot needs to localize on a known map using only tactile signals.
Various techniques are available to approximate discrete domains with continuous ones. In this thesis, we aim to experiment with these techniques and tailor them for tactile measurements to facilitate gradient-based control. This will involve
Time permitting, we could extend this work by pursuing several avenues, such as full tactile SLAM, alternative tasks based on tactile measurements, and comparing different sample-based methods for contact-seeking motion planning.
You can find all the necessary information here.